Function build_pod_labels_from_instance 

pub fn build_pod_labels_from_instance(
    instance_name: &str,
    instance: &Bind9Instance,
) -> BTreeMap<String, String>

Builds the label set stamped onto the Pods of a Bind9Instance.

This is deliberately a superset of build_labels_from_instance, which remains the Deployment’s spec.selector and the Service’s selector.

Why the two are separate

spec.selector on a Deployment is immutable — Kubernetes rejects any change to it, because changing which Pods a Deployment claims would orphan the ones it used to own. When one label map fed both the selector and the Pod template (as it did before topology spreading landed), adding any new label to Pods would have changed the selector too, wedging the reconciler on every Deployment that already existed.

So: build_labels_from_instance is frozen and owns the selector, and everything added afterwards goes here. Kubernetes only requires that the selector matches the template labels, so the template may carry extras. Service selectors are subset matches and are unaffected.

Extra labels

• bindy.firestoned.io/cluster — the owning cluster, taken from spec.clusterRef. Without it there is no label shared by the sibling single-Pod Deployments of a cluster, and so no way to write a topology spread selector that balances all primaries against each other.
• bindy.firestoned.io/role — derived from spec.role rather than from the CR’s metadata, so it is present even on a hand-written Bind9Instance that carries no role label. Only inserted when the selector does not already carry the key, so the Pod can never stop matching its own Deployment’s selector.